The aim of this study was to develop a novel cognitive procedure for operationalizing how the re-encoding and constraint relaxation, suggested by representational change theory (RCT) (Ohlsson, 1992, 2011), can effect representational change in verbal insight problem solving, thus circumventing the constraints imposed by past experience. Some participants were trained in using an evaluative cognitive procedure that aimed to facilitate the identification of any inconsistency between the participant's interpretation of the problem and the problem statement, and thus cue the re-encoding proposed by RCT. In Experiment 1, participants were randomly allocated to training, practice, or a no-training control condition, and were subsequently tested on 7 verbal insight problems. Concurrent verbal protocols were collected and analyzed to identify problem solvers' proposed hypotheses and also to assess whether problem solving behavior changed in line with the training. Inconsistency identification training, rather than practice or no training, improved solution rate across novel problems and resulted in more paraphrasing and questioning of the problem statement, and a modest increase in participants' reflection on their problem solving. Results from Experiment 2 indicated that this improvement in representation change through training was not due to increased awareness of the nature of verbal insight problems but rather training in identifying inconsistencies between the problem statement and a person's interpretation of it. Experiment 3 revealed that the performance improvement with training was sustained after a delay of 48 hr. Theoretical and methodological issues are discussed.